AI Features
AI Features
Information and guidance on AI features including security and responsible use
This page provides an overview of some of the most commonly used AI features, what they can be used for, and key considerations for security, privacy, and responsible use.
- The guidance on this page applies only to Harvard-approved AI tools. A list of currently approved tools is available here.
- If you need access to a feature that is not currently enabled, would like to request approval for a new AI tool or feature, or would like additional guidance on the use of AI features, contact HUIT.
- It's important to note that AI tools are constantly changing and adding new features. We will add more features to this page, and you can find more detailed information about specific features on each vendor's specific product website.
Jump to feature
đź“‹ Agents
AI agents are a feature of some AI tools that can carry out multi-step tasks on your behalf, rather than simply responding to a single prompt. You describe the outcome you want, and the agent works through the task by deciding what action to take next—such as searching for information, opening files, running code, or editing documents—then evaluating the result before continuing. It repeats this process until the task is complete, without requiring a new prompt at every step.
AI agents can be useful for work involving several steps, such as research, coding, or drafting and revising documents. Many agents can work in the background while you focus on other tasks.
Understand what the agent can access
AI agents may be able to search files and messages, use tools on your device, run commands, and take actions on your behalf.
- Before using an agent, understand what data and systems it can access.
- Review the permissions it requires and the actions it can perform.
- Only use approved tools and limit the agent’s access to Level 3 data or below.
Review the agent's outputs
As with any AI-generated content, agent-generated work may contain errors or incomplete information.
- Review outputs before sharing, submitting, or relying on them.
- Do not allow an agent to send messages, modify records, change permissions, submit forms, deploy code, or trigger workflows without careful human review.
Review the agent's actions
Because AI agents make decisions throughout a multi-step task, mistakes can compound if your instructions are incomplete or misunderstood.
- Review both the final result and the actions the agent took to complete the task.
- Confirm that the agent completed only the actions you intended.
- Check an agent's progress periodically, especially during long-running tasks, and stop the task if the agent begins producing incorrect or unintended results.
🧩 Connectors
AI connectors allow AI tools to access and integrate with other systems or add new capabilities. For example, you can connect tools such as ChatGPT or Claude to your email, calendar, files, cloud storage, collaboration tools, code repositories, and other apps. This allows the AI tool to use the data and functionality of connected apps to retrieve information, answer questions, and complete tasks on your behalf.
đź’» Coding
AI coding tools can help generate, explain, debug, test, review, and document code. Some tools can also connect to repositories, create pull requests, run commands, or support software development workflows.
AI coding tools can make software development faster, but AI-generated code still requires human review. Developers remain responsible for the security, privacy, accessibility, licensing, quality, and maintainability of any code they use.
Review and test AI-generated code
AI-generated code can contain errors, introduce security vulnerabilities, use outdated libraries or patterns, misunderstand your intended system or data flow, or include inappropriate dependencies.
- Review, test, and understand AI-generated code before using it.
- Follow your normal code review, testing, and change management practices before publishing, merging, or deploying code.
- Be especially careful with code that affects authentication, authorization, encryption, logging, data handling, infrastructure, or production systems.
Protect sensitive information
Code and related technical information can contain confidential or security-sensitive data.
- Do not enter credentials, secrets, private keys, tokens, or certificates into AI tools.
- Do not enter confidential source code, sensitive logs, production configuration files, or restricted system details into unapproved tools.
- Whenever possible, use test data instead of production data.
Understand what the tool can access and do
Some AI coding tools can connect to repositories, create branches or pull requests, run commands, or interact with CI/CD systems.
- Before using these capabilities, understand what repositories, systems, and data the tool can access.
- Ask for help before using an AI coding tool if it connects to private repositories, can make changes on your behalf, interacts with CI/CD systems, involves confidential or regulated data, affects production systems, or has not been approved for your intended use.
đź“ť Meeting Assistants
AI meeting assistants, sometimes called AI note takers or meeting bots, can record, transcribe, summarize, or analyze online meetings.
These tools can improve accessibility and make it easier to capture notes and action items. However, they can also pose privacy, regulatory, and legal risks, and have the potential to stifle conversation and open inquiry.
At this time, AI meeting assistants should not be in Harvard meetings with the exception of approved tools as part of limited HUIT-directed pilot programs or approved accessibility accommodations.